Fujitsu upgrades RIKEN's AI supercomputer
Fujitsu has announced an upgrade to RAIDEN (Riken AIp Deep learning ENvironment), a system for artificial intelligence research originally deployed in 2017 to the RIKEN Center for Advanced Intelligence Project (AIP Center), the AI research arm of RIKEN.
The upgraded RAIDEN has increased its performance considerably, moving from an initial total theoretical performance of four petaflops, to 54 PFLOPS. The system upgrade, which goes into operation from April 2018, will support RIKEN’s AI research taking advantage of the new performance and upgraded GPUs.
Since it began operations after system delivery in April 2017, the RIKEN AIP Center in Japan has put RAIDEN to use for research on next generation AI technology. This includes AI research conducted with enormous neural networks, a machine learning method particularly in deep learning. The increasing scale of neural networks provides scope for increased accuracy and more complex characteristics to be studied but has also led to a drastic increase in computational requirements.
The RIKEN AIP Center undertook this upgrade as usage and potential research requirements continue to grow. This new upgrade will allow the centre to increase the efficiency of research and development and promote further AI research using RAIDEN.
The upgraded system uses the 32 existing Fujitsu Server PRIMERGY RX2530 M2 x86 servers in addition to 64 new Fujitsu PRIMERGY CX2550 M4 servers. The system also includes one Fujitsu Server PRIMERGY RX4770 M4 unit as a compute server that handles high-volume data.
For the GPU servers specialised for deep learning, the system has been upgraded from NVIDIA DGX-1 servers featuring NVIDIA Tesla P100 GPUs to the latest NVIDIA Tesla V100 GPUs. In addition, by increasing the number of DGX-1 servers to 54, the upgraded system now achieves 54 PFLOPS.